MDST-DGCN: A Multilevel Dynamic Spatiotemporal Directed Graph Convolutional Network for Pedestrian Trajectory Prediction

Table 2

ADEs and FDEs of different methods for long-term trajectory prediction on the PEDWALK with various pedestrian densities.

Density (d)	10 ≤ d ≤ 30	30 ≤ d ≤ 50	50 ≤ d ≤ 70	70 ≤ d ≤ +∞	Overall
SGAN	35.57/70.39	44.02/87.08	43.30/85.84	47.34/93.24	44.02/86.96
SGAN-P	36.06/71.02	41.92/81.39	40.70/78.70	45.09/87.39	42.03/81.54
STGAT	33.20/60.21	38.06/68.25	38.33/69.18	41.97/75.98	39.02/70.47
MDST-DGCN-D	32.62/63.05	36.38/69.15	35.61/67.17	40.80/77.77	37.31/70.80
MDST-DGCN-S	30.53/57.88	34.62/64.81	34.68/64.81	39.75/75.21	35.99/67.53

The density (d) means the number of pedestrians in the scenario, and D1 ≤ d ≤ D2 means we select the samples in which the number of pedestrians is not less than D1 and not greater than D2. All methods predict 9.6 seconds, given the previous 6.4 seconds. Errors reported are ADE/FDE in pixels on the original size of 1920 × 1080. Methods marked with draw 20 samples and select the best sample. The values with the least error in the comparison model are bolded.